InfoResearchPreprintLLM-specific
Intent Contracts: Taking Tools Away from Agents
- Published
- Record updated
Summary
Intent contracts are an authorization interface for tool-using LLM agents, in which the planning model holds no tools and submits a logged contract per step. A trusted model checks the contract against the user's words and writes the tool call. Contracts completed the user's task in 27/30 runs with planner-added work, versus 16/30 for an intent-visible reviewer, but they do not reduce unauthorized actions, and wrong-target errors occurred in 40-49% of runs.
Related items
- CriticalGHSA-9x8x-crjq-g64h: Jupyter MCP Server: Browser DNS rebinding can hijack the Jupyter backend through management routesSimilar attack · GitHub Advisory Database
- HighGHSA-24pw-rf5p-mgp7: Jupyter MCP Server: Unauthenticated management routes can hijack the Jupyter backend in standalone HTTP modeSimilar attack · GitHub Advisory Database
- HighGHSA-m932-63jc-4jqq: k8s-mcp-server: Missing authentication on default SSE/HTTP transport allows unauthenticated Kubernetes tool executionSimilar attack · GitHub Advisory Database
- HighGHSA-3mgh-93r5-r7jf: CodeWhale: exec_shell parallel-readonly classifier auto-approves fd/rg/git grep exec flags enabling silent RCESimilar attack · GitHub Advisory Database
- HighGHSA-qx49-fqc8-xw99: Model Context Protocol Python SDK: OAuth client could send credentials to an authorization server chosen by the MCP serverSimilar attack · GitHub Advisory Database